Papers with policy optimization
Guided Dialog Policy Learning: Reward Estimation for Multi-Domain Task-Oriented Dialog (D19-1)
Copied to clipboard
| Challenge: | Existing methods to learn dialog policy require elaborate design and user goals. |
| Approach: | They propose an algorithm that estimates the reward signal and infers the user goal in dialog sessions. |
| Outcome: | The proposed algorithm achieves higher task success than state-of-the-art models on a multi-domain task-oriented dialog dataset. |
AVAST: Attentive Variational State Tracker in a Reinforced Navigator (2022.aacl-main)
Copied to clipboard
| Challenge: | Recent advances in reinforcement learning have been proposed to deal with robotic navigation problems, especially vision-and-language navigation task. |
| Approach: | They propose a method to approximate belief state distribution for the construction of a reinforced navigator by using a variational approach to approximate the unseen environment. |
| Outcome: | The proposed method improves generalization to the unseen environment which is barely achieved by traditional deterministic state tracker. |
Unearthing Gems from Stones: Policy Optimization with Negative Sample Augmentation for LLM Reasoning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Recent advances in reasoning language models have witnessed a paradigm shift from short to long CoT pattern. |
| Approach: | They propose a behavior-constrained policy gradient with negative sample augmented (BCPG-NSA) negative steps are valuable components in long CoT models, authors argue . |
| Outcome: | The proposed framework outperforms baselines on math/coding reasoning benchmarks using the same training dataset. |
Fine-Tuning Language Models with Reward Learning on Policy (2024.naacl-long)
Copied to clipboard
| Challenge: | Reinforcement learning from human feedback (RLHF) is an effective approach to align large language models (LLMs) to human preferences. |
| Approach: | They propose a framework that refines a reward model using policy samples to keep it on-distribution. |
| Outcome: | The proposed framework outperforms the state-of-the-art on three benchmark datasets showing that it can learn robust representations of policy samples. |
Scheduled Dialog Policy Learning: An Automatic Curriculum Learning Framework for Task-oriented Dialog System (2021.findings-acl)
Copied to clipboard
| Challenge: | et al., 2013) show that dialog policy learning is an important component of the task-oriented dialogue system. |
| Approach: | They propose a framework that integrates curriculum learning and policy optimization . they propose to train dialog agents from easy dialogues to complex ones . |
| Outcome: | The proposed framework outperforms the state-of-the-art model on multi-task dialogues. |
TemplateRL: Structured Template-Guided Reinforcement Learning for LLM Reasoning (2026.findings-acl)
Copied to clipboard
Jinyang Wu, Chonghua Liao, Mingkuan Feng, Shuai Zhang, Zhengqi Wen, Haoran Luo, Ling Yang, Huazhe Xu, Jianhua Tao
| Challenge: | Existing RL methods rely on unstructured self-sampling to fit scalar rewards, resulting in inefficient rollouts. |
| Approach: | They propose a structured template-guided RL framework that augments policy optimization with explicit template guidance. |
| Outcome: | Experiments show that TemplateRL outperforms GRPO and GRPI by 99% on AIME and 41% on AMC with superior stability on weak models and remarkable cross-domain generalization. |
Reward Mixology: Crafting Hybrid Signals for Reinforcement Learning Driven In-Context Learning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for in-context learning (ICL) performance rely on quality and ordering of demonstrations. |
| Approach: | They propose a method that models iterative demonstration selection as a Markov Decision Process and craft hybrid reward signals. |
| Outcome: | The proposed method combines outcome-based accuracy signals with process-oriented signals like stepwise influence and label entropy improvement. |
Agent-Pro: Learning to Evolve via Policy-Level Reflection and Optimization (2024.acl-long)
Copied to clipboard
Wenqi Zhang, Ke Tang, Hai Wu, Mengna Wang, Yongliang Shen, Guiyang Hou, Zeqi Tan, Peng Li, Yueting Zhuang, Weiming Lu
| Challenge: | Large Language Models (LLMs) are designed as specific task solvers with sophisticated prompt engineering, but are inherently incapacitating to address complex dynamic scenarios. |
| Approach: | They propose an LLM-based agent with policy-level reflection and optimization that can learn from interactive experiences and progressively elevate its behavioral policy. |
| Outcome: | The proposed agent outperforms vanilla LLM and specialized models in blackjack and Texas hold’em. |
Visually-Guided Policy Optimization for Multimodal Reasoning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing RLVRs lack visual faithfulness due to text-dominated reasoning . a novel framework to reinforce visual focus during policy optimization is proposed . |
| Approach: | They propose a framework to reinforce visual focus during policy optimization using visual attention compensation mechanism. |
| Outcome: | The proposed framework exhibits better visual activation and superior performance in multimodal reasoning and visual-dependent tasks. |
DDxGym: Online Transformer Policies in a Knowledge Graph Based Natural Language Environment (2024.lrec-main)
Copied to clipboard
Benjamin Winter, Alexei Gustavo Figueroa Rosero, Alexander Loeser, Felix Alexander Gers, Nancy Katerina Figueroa Rosero, Ralf Krestel
| Challenge: | specialized OpenAI Gym environment for clinical differential diagnosis is limited by data access due to privacy concerns. |
| Approach: | They propose a specialized OpenAI Gym environment for clinical differential diagnosis . they frame the task as a natural-language-based reinforcement learning problem . |
| Outcome: | The proposed model improves over baselines and improves on existing models. |
Learning to Retrieve Iteratively for In-Context Learning (2024.emnlp-main)
Copied to clipboard
Yunmo Chen, Tongfei Chen, Harsh Jhamtani, Patrick Xia, Richard Shin, Jason Eisner, Benjamin Van Durme
| Challenge: | In-context learning is a powerful tool for learning large language models. |
| Approach: | They propose an iterative retrieval framework that empowers retrievers to make iterable decisions through policy optimization. |
| Outcome: | The proposed framework outperforms existing methods on semantic parsing datasets with 4M additional parameters for state encoding. |
GA-S3: Comprehensive Social Network Simulation with Group Agents (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing social network simulations focus on discrete events or system dynamics instead of elucidating underlying mechanisms or causal relationships. |
| Approach: | They propose a Social network simulation system that leverages newly designed Group Agents to make intelligent decisions regarding various online events. |
| Outcome: | The proposed system can make intelligent decisions regarding online events at a manageable cost. |
Beyond Pedagogical Principles: Multi-Horizon Preference Optimization for Efficient Socratic Tutoring (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for developing LLMs are constrained by static data or sparse reward signals in online settings. |
| Approach: | They propose a framework that iteratively refines tutor agents using a multi-horizon reward function within a dynamic teacher-student simulation environment. |
| Outcome: | The proposed framework improves model performance and balances principles and effectiveness compared to baselines. |
Provably Safe Offline-to-Online RL: Decoupling Learning from Data-Driven Safety Enforcement (2026.acl-long)
Copied to clipboard
| Challenge: | Hybrid offline–online reinforcement learning (O2O RL) promises both sample efficiency and robust exploration, but suffers from instability due to distribution shift between offline and online data. |
| Approach: | They propose a framework that decouples policy optimization from safety enforcement . they propose dynamic curricula that gradually extend temporal horizons and anneal offline–online data mixing . |
| Outcome: | The proposed framework preserves the exploratory value of online interactions without collapsing to conservative policies. |
Not All Tokens Matter: Towards Efficient LLM Reasoning via Token Significance in Reinforcement Learning (2026.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) often produce unnecessarily long explanations that reduce efficiency. |
| Approach: | They propose a length-aware reward that selectively penalizes insignificance tokens . they also propose 'dynamic length control' that encourages more detailed reasoning . |
| Outcome: | The proposed method reduces response length while maintaining correctness, the authors show . it selectively penalizes insignificance tokens while maintaining accuracy . |
Free Energy-Driven Reinforcement Learning with Adaptive Advantage Shaping for Unsupervised Reasoning in LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Existing unsupervised reinforcement learning methods lack the capacity to adapt to the model’s evolving reasoning capabilities during training. |
| Approach: | They propose an unsupervised reinforcement learning algorithm that adapts rewards to balance consensus and exploration based on the Free Energy Principle. |
| Outcome: | Empirical evaluations on nine datasets show that FREIA outperforms baseline methods on reasoning tasks. |
VRPO: Rethinking Value Modeling for Robust RL under Noisy Supervision in LLM Post-Training (2026.acl-long)
Copied to clipboard
Dingwei Zhu, Shihan Dou, Zhiheng Xi, Senjie Jin, Guoqiang Zhang, Jiazheng Zhang, Junjie Ye, Mingxu Chai, Enyu Zhou, Ming Zhang, Yuhui Wang, Caishuang Huang, Chenhao Huang, Yunke Zhang, Yuran Wang, Tao Gui, Qi Zhang, Xipeng Qiu, Xuanjing Huang
| Challenge: | Reinforcement Learning (RL) in real-world environments often suffers from ambiguous or incomplete supervision. |
| Approach: | They propose a framework that enhances value modeling for robust RL in LLM post-training by integrating auxiliary losses guided by entropy and perplexity from a frozen language model and variational information bottleneck. |
| Outcome: | The proposed framework outperforms baselines on multi-turn dialogue, math reasoning, and science QA with rule-based and model-based rewards. |
ERRV: Eliciting Efficient Reasoning through Reasoning Vectors for Policy Optimization in Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing efforts to improve reasoning efficiency of large language models focus on modifying the reinforcement learning reward, such as adding length penalties. |
| Approach: | They propose a training framework that elicits efficient reasoning through reasoning vectors and a framework that allows the model to generate high-quality responses during reinforcement learning. |
| Outcome: | The proposed framework reduces reasoning length by 30% while maintaining stability, while retaining high accuracy. |
Enhancing RLHF with Human Gaze Modeling (2025.emnlp-main)
Copied to clipboard
| Challenge: | Reinforcement Learning from Human Feedback (RLHF) is a powerful paradigm for aligning language models with human values and preferences. |
| Approach: | They propose to use gaze-aware reward models and gaze-based distribution of sparse rewards to enhance RLHF. |
| Outcome: | The proposed models achieve faster convergence while maintaining or slightly improving performance, reducing computational requirements during policy training. |
Verifier-Free RL for LLMs via Intrinsic Gradient-Norm Reward (2026.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have demonstrated remarkable capabilities in reasoning tasks and general tasks. |
| Approach: | They propose a "Verifier-free Intrinsic Gradient-Norm Reward" that uses only the policy model itself. |
| Outcome: | The proposed reward outperforms the state-of-the-art RLIF baseline INTUITOR on math benchmarks and shows cross-domain transfer to code benchmarks when trained only on math data. |
Can Compact Language Models Search Like Agents? Distillation-Guided Policy Optimization for Preserving Agentic RAG Capabilities (2026.acl-long)
Copied to clipboard
| Challenge: | Existing agentic RAG systems rely on large language models with billions of parameters. |
| Approach: | They propose a method to elicit agentic RAG behaviors from compact models . they propose ARC, which uses cold-start initialization and teacher guidance . |
| Outcome: | The proposed method outperforms the larger teacher model in some cases. |
Foresight Optimization for Strategic Reasoning in Large Language Models (2026.acl-long)
Copied to clipboard
Jessie Wang, Jiawen Duan, Jian Wang, Kaitao Song, Chunpu Xu, Johnny K. W. Ho, YU Fenggang, Johan F. Hoorn, Wenjie Li
| Challenge: | Existing reasoning enhancement methods do not capture foresight in LLMs. |
| Approach: | They propose to integrate opponent modeling principles into policy optimization to enhance strategic reasoning in LLMs by integrating opponent modeling into policy. |
| Outcome: | The proposed method outperforms existing reasoning-based LLMs in out-of-domain scenarios and shows that it significantly enhances strategic reasoning across LLM of varying sizes and origins. |
Explanation Quality Assessment as Ranking with Listwise Rewards (2026.findings-acl)
Copied to clipboard
| Challenge: | a new approach to explanation quality assessment is to rank explanations by relative quality . standard reward objectives do not preserve graded distinctions well enough for policy optimization . |
| Approach: | They reformulate explanation quality assessment as a ranking problem instead of a generation problem . they train listwise and pairwise ranking models to preserve ordinal structure . |
| Outcome: | The proposed model outperforms regression on score separation and performance on listwise and pairwise models. |
DARM: Distribution-Aware Reward Modeling by Alleviating Biases from Low Preference-Context Dependency Data (2026.acl-long)
Copied to clipboard
Shaofan Liu, Guoqiang Zhang, Shihan Dou, Huiyuan Zheng, Yiming Zhou, Junjie Ye, Shaowen Wang, Shichun Liu, Jiazheng Zhang, Tao Gui, Qi Zhang, Xuanjing Huang
| Challenge: | Existing methods for training reward models are vulnerable to context neglect and degraded accuracy. |
| Approach: | They propose distribution-aware reward modeling that augments the RM objective with a conditional mutual information regularizer that maximizes context and the predicted reward conditioned on the response. |
| Outcome: | The proposed model improves performance in RLHF and improves accuracy in other settings. |
Influence-based Online Experience Selection for Effective RLHF (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for RL fail to establish an interpretable connection between data and optimization objectives. |
| Approach: | They propose a data selection method that dynamically estimates the influence of individual training samples on policy optimization. |
| Outcome: | The proposed method significantly improves training effectiveness with fewer optimization steps. |